| name | arib-dev-lean |
| argument-hint | <path | diff> |
| description | Dev | Over-engineering review — reads a diff/module and returns a delete-list of bloat (single-use abstractions, premature generalization, speculative config, dead options) WITHOUT stripping legitimate structure. Advisory: returns recommendations, never auto-deletes. The on-demand companion to the ponytail-lite tripwire hook. Authored natively (no Ponytail dependency). ADR-033. |
/arib-dev-lean — review for over-engineering (YAGNI)
Read a diff or module and return a delete-list: the bloat that can go without losing
function. The honest counterpart to the ponytail-lite.sh tripwire — that hook is a quiet
edit-time hint; this is the deliberate review. Advisory only — it returns
recommendations; it never deletes anything itself.
Authored natively (CCM's own YAGNI ladder) — the developer plan proposed wrapping the
Ponytail tool, which isn't installed; this needs no external dependency.
What it flags (the YAGNI ladder)
- Single-use abstraction — an interface/abstract class/factory/strategy with exactly
one implementer and one caller. Inline it until a second user exists.
- Premature generalization — a "configurable" function whose every call passes the
same options; a generic the codebase only instantiates one way.
- Speculative config / flags — options, env vars, or feature flags that nothing reads,
or that have exactly one value everywhere.
- Dead options & params — parameters always passed the same value; branches never taken.
- Indirection with no payoff — a pass-through wrapper/adapter that only forwards.
- Gold-plated error handling — catch-rethrow that adds nothing; defensive checks for
states the type system already prevents.
What it MUST NOT flag (legitimate structure wins — CCM rule)
- Framework structure required by the stack — NestJS modules/providers/guards/DTOs,
React boundaries, etc. (the
ponytail-lite hook exempts *.module/*.controller/*.service/ *.guard/*.dto; this review honors the same intent).
- An abstraction with a real second user or a documented near-term one.
- Anything marked
// ccm-ceremony: — intentional scaffolding.
- Security/validation depth on a high-stakes path — never "lean away" a guard.
When in doubt, keep it and note it as "watch" rather than "delete". A false delete is
worse than a surviving abstraction.
Protocol
- Read the target (a path, a module, or
git diff <base>...<head>).
- Walk the ladder above; for each hit, record: file:line · what · why it's single-use ·
the concrete simplification · a confidence (high/med/low).
- Skip the exemptions; downgrade anything uncertain to "watch".
- Output a delete-list (high-confidence) + a watch-list (uncertain) →
io/ledger/lean-review-<date>.md (standard CCM audit format). Recommend; do not apply.
- Optionally hand the delete-list to
/arib-build or a specialist to action — behind the
normal merge gate.
Anti-patterns
Stripping required framework ceremony · deleting a guard/validation to "simplify" ·
auto-applying deletes (this is advisory) · flagging an abstraction that has a real second
user · treating low-confidence hits as deletes.
